A Biogeographical Regionalization of Angolan Mammals Patrícia RODRIGUES Instituto De Investigação Cientíﬁca Tropical, R

bs_bs_banner

Mammal Review ISSN 0305-1838

REVIEW A biogeographical regionalization of Angolan mammals Patrícia RODRIGUES Instituto de Investigação Científica Tropical, R. da Junqueira, 86 – 1°, 1300-344 Lisboa, Portugal, and CIBIO/InBio – Centro de Investigação em Biodiversidade e Recursos Genéticos, Universidade do Porto, Campus Agrário de Vairão, 4485-661 Vairão, Portugal. E-mail: [email protected] Rui FIGUEIRA Instituto de Investigação Científica Tropical, R. da Junqueira, 86 – 1°, 1300-344 Lisboa, Portugal, and CIBIO/InBio – Centro de Investigação em Biodiversidade e Recursos Genéticos, Universidade do Porto, Campus Agrário de Vairão, 4485-661 Vairão, Portugal. E-mail: rui.fi[email protected] Pedro VAZ PINTO CIBIO/InBio – Centro de Investigação em Biodiversidade e Recursos Genéticos, Universidade do Porto, Campus Agrário de Vairão, 4485-661 Vairão, Portugal, Departamento de Biologia, Faculdade de Ciências, Universidade do Porto, Rua do Campo Alegre s/n. 4169-007 Porto, Portugal, ISCED – Instituto Superior de Ciências da Educação da Huíla, Rua Sarmento Rodrigues, Lubango, Angola, and The Kissama Foundation, Rua Joaquim Capango n°49, 1°D, Luanda, Angola. E-mail: [email protected] Miguel B. ARAÚJO Departmento de Biogeografía y Cambio Global, Museo Nacional de Ciencias Naturales, CSIC, Calle José Gutiérrez Abascal, 2, 28006 Madrid, Spain, CIBIO/InBio, Universidade de Évora, Largo dos Colegiais, 7000 Évora, Portugal, and Center for Macroecology, Evolution and Climate, Natural History Museum of Denmark, University of Copenhagen, Universitetsparken 15, DK-2100 Copenhagen, Denmark. E-mail: [email protected] Pedro BEJA* CIBIO/InBIO – Centro de Investigação em Biodiversidade e Recursos Genéticos, Universidade do Porto, Campus Agrário de Vairão, 4485-661 Vairão, Portugal, Departamento de Biologia, Faculdade de Ciências, Universidade do Porto, Rua do Campo Alegre s/n. 4169-007 Porto, Portugal, and ISCED – Instituto Superior de Ciências da Educação da Huíla, Rua Sarmento Rodrigues, Lubango, Angola. E-mail: [email protected]

Keywords ABSTRACT environmental gradient, historical distribution, South-West Africa, vertebrates, zoogeography 1. We developed a biogeographical regionalization of Angolan mammals based on data collected before major declines occurred during the civil war (1975– *Correspondence author. 2002). In terms of its biodiversity, Angola is one of the least known of all African countries. Submitted: 2 July 2014 2. We used 9880 grid records of 140 species (rodents, ungulates and carnivores) Returned for revision: 11 November 2014 Revision accepted: 15 January 2015 collected mainly in 1930–80, at a quarter degree cell resolution. Biogeographical β Editor: KH regions were identiﬁed by using cluster analysis, based on sim dissimilarity matri- ces and a hierarchical classiﬁcation using Ward’s method. An indicator value doi:10.1111/mam.12036 analysis was used to identify species characterizing each region. Distance-based redundancy analysis was used to investigate the environmental correlates of mammalian assemblages. 3. Four biogeographical subdivisions emerged from ungulate distributions, while rodent and carnivore data were largely uninformative. In the north, the Zaire- Lunda-Cuanza region was mainly characterized by ungulate species associated with Congolian forests. In the south, the Namibe and Cunene-Cuando Cubango regions were mainly characterized by ungulates widespread in south-western and southern Africa. In between these regions, the Central Plateau region was mainly

characterized by a few widespread ungulate species that are relatively common in dense miombo woodlands. 4. Biogeographical patterns were significantly associated with a dominant north– south gradient of decreasing humidity and increasing temperature, and with a concurrent gradient from dense forests and woodlands to open savannas, grasslands and deserts. 5. The biogeographical regions we identified in Angola were largely consistent with other bioregionalizations developed using various taxonomic groups at larger spatial scales. Biogeographical patterns reflected the southward penetration of Congolian forest species in the north, and the northward penetration of southern African desert/grassland species in the south-west and of open savanna species in the south. These processes seem to be controlled by the distribution of vegetation types, which in turn are associated with climatic gradients and soil types. The stronger patterns observed for ungulates than for other mammals may reflect the close association of ungulates to specific vegetation types.

INTRODUCTION amphibians (Penner et al. 2011), birds (de Klerk et al. 2002, Muñoz et al. 2003) and mammals (Gelderblom et al. 1995, Biogeographical regionalizations allow the definition of Rowe-Rowe & Taylor 1996, Grubb et al. 2000). Researchers homogeneous regions in terms of species assemblages, and analysing both plant and animal distributions proposed the the identification of factors potentially shaping the spatial division of Sub-Saharan Africa into the Saharan, Sudanian, distribution of such assemblages. Therefore, they are Ethiopian, Somalian, Congolian, Zambezian and Southern increasingly regarded as a preliminary step towards the African regions (Linder et al. 2012). These broad biogeo- planning of conservation efforts (Ladle & Whittaker 2011). graphical regions, however, were recognized as containing Despite their value, development of accurate biogeographi- substantial internal structure, corresponding to sub-regions cal regionalizations is limited by the shortage of informa- characterized, for instance, by particular environmental con- tion on species’ distributions, particularly in poorly ditions or by the presence of centres of endemism and islands explored regions of the world. In such areas, species’ distri- of high diversity (Linder et al. 2012). Accurate identification bution data often reflect the distribution of researchers’ of sub-regions would require analyses conducted at finer interests rather than that of species, creating a number of resolution than typically carried out; such analyses could spatial biases that may confound biological patterns (Pyke strongly contribute to improving our knowledge of African & Ehrlich 2010). Furthermore, the time frame for data col- biogeography. lection may be considerably different from the time frame In this study, we undertake a biogeographical regio- of the analysis, thus bringing in additional inaccuracies with nalization of Angola based on previously unavailable data regard to the current situation (Pyke & Ehrlich 2010). The on the historical distribution of mammalian species. Angola practical application of these studies may also be limited is a large and biodiverse country in south-western Africa, because data resolution may be too coarse to delineate accu- which includes areas of global conservation importance rately the boundaries between adjacent regions. Develop- such as the coastal escarpment, but where biodiversity ment of biogeographical regionalizations in poorly known research is largely lacking (Myers et al. 2000, Figueiredo areas thus remains a theme of major applied value, which et al. 2009, Mills 2010, Romeiras et al. 2014). Linder et al. may help to validate studies conducted at larger spatial (2012) classified most of the country as falling into the scales and contribute towards refining the prioritization of Zambezian region, with a transition fringe in the Congolian conservation action. biogeographical unit in the north (Shaba), and transition Efforts to develop comprehensive biogeographical regio- fringes in the South African unit in the south-west (south- nalizations for Africa have a long history. Early researchers western Angola) and in the south (Kalahari). The boundar- investigated the biogeographical patterns of the plants ies between units, however, were highly sketchy and varied (Lebrun 1947, Monod 1957, White 1965), butterflies to some extent depending on the taxonomic group (Linder (Carcasson 1964), amphibians (Poynton 1964) and birds et al. 2012). Furthermore, within the large Zambezian (Chapin 1932) of Sub-Saharan and Southern Africa.Research region, there is probably much biotic heterogeneity, as sug- on this topic has proceeded during the past two decades, and gested for instance by the identification of several vegetation biogeographical regionalizations have been proposed for units in early phytogeographic research (Barbosa 1970) and

104 Mammal Review 45 (2015) 103–116 © 2015 The Mammal Society and John Wiley & Sons Ltd P. Rodrigues et al. Biogeographical regionalization of Angolan mammals by high levels of species turnover (Linder et al. 2012). The Congo analysis of mammalian data could contribute towards resolving these uncertainties, as mammals are a relatively Democratic specious group for which a wealth of information exists Republic of Congo from colonial times (mainly from the 20th century; Linder et al. 2012). Although humans have a strong impact on mammalian distributions, particularly on those of large carnivores and ungulates, most of the data we used were Luanda collected before the civil war (1975–2002) when widespread poaching strongly affected a range of species (Crawford-Cabral & Veríssimo 2005, Chase & Grifﬁn 2011). The aims of this study were to: (1) review and compile in a digital format the data on mammalian species’ distributions for Angola; (2) use the data to identify biogeographical units or regions, based on repeatable statistical procedures; (3) identify ‘indicator’ species that discriminate the different regions; and (4) identify the environmental correlates of the biogeographical patterns uncovered. Results were interpreted in the light of current understand- Zambia ing of the biogeography of this region of Africa.

METHODS Namibia Botswana Study area × 6 2 Angola is a large country (1.25 10 km ) located on the AECFor A MioWd KaokDes south-western coast of Africa (4°22′S, 18°03′S; Fig. 1). CAMang A MFor-Gra Nb SavWd Elevation varies from 0 to 2620 m above sea level; the litto- WC ForSav CZ MioWd A MopWd ral plains are separated from the inland plateau by a north– SC ForSav ZCy DryFor ZBai Wd south escarpment running parallel to the coast. Climate AScarp SWd WZ Gras ZFlood Gra ranges from tropical wet/humid in the north to extremely arid in the south-west. Angola has a diverse array of ecosys- Fig. 1. Geographical location of Angola, Africa, showing its major tems and is biologically mega-diverse, encompassing 15 rivers and WWF Ecoregions: AECFor, Atlantic Equatorial Coastal Forest; World Wildlife Fund (WWF) Ecoregions (Olson et al. 2001; CAMang, Central African Mangroves; WCForSav, Western Congolian Forest-Savanna Mosaic; SCForSav, Southern Congolian Forest-Savanna Fig. 1). In the north and north-west, there are Congolian Mosaic; AScarpSwd, Angolan Scarp Savanna and Woodlands; forest-savanna mosaics and the scarp savannas and wood- AMioWd, Angolan Miombo Woodlands; AMFor-Gra, Angolan lands. The south-western coastal plain is occupied by the Montane Forest-Grassland Mosaic; CZMioWd, Central Zambezian Kaokoveld desert and Namibian dry savanna woodlands. Miombo woodlands; ZCyDryFor, Zambezian Cryptosepalum Dry The most widespread vegetation is the miombo woodland Forests; WZGras, western Zambezian Grasslands; KaokDes, Kaokoveld occupying the Central Plateau. In the southern borders, the Desert; NbSavWd, Namibian Savanna Woodlands; AMopWd, Angolan miombo is replaced by mopane Colophospermum mopane Mopane Woodlands; ZBaiWd, Zambezian Baikiaea Woodlands; ZFloodGra, Zambezian Flooded Grasslands. and baikiaea Baikiaea plurijuga woodlands. The terrestrial mammalian fauna comprises 274 species, 13 of which are endemic and 19 are globally threatened (Wilson & Reeder 2005, Anonymous 2013). During the civil war, there was Cabral & Veríssimo 2005), we compiled and georeferenced, uncontrolled poaching of large mammals, so their current at a resolution of quarter degree grid cell (approximately status is largely unknown (Crawford-Cabral & Veríssimo 25 × 25 km), a total of 9880 occurrence records of 140 2005, Pitra et al. 2006, Chase & Grifﬁn 2011). rodent, ungulate and carnivore species (Fig. S1 and Table S1 in Appendix S1). The data set is available for download at http://www.gbif.org/dataset/2fea4042-5aba-4cda-bab9-adcf Species data a2bbe97d. Most information was collected from 1930 to Based on data available in the literature (Crawford-Cabral & 1980, but the oldest records date back to 1860 (Fig. S2 in Simões 1987, 1988, Crawford-Cabral 1998, Crawford- Appendix S1). The north-western enclave of Cabinda was

Mammal Review 45 (2015) 103–116 © 2015 The Mammal Society and John Wiley & Sons Ltd 105 Biogeographical regionalization of Angolan mammals P. Rodrigues et al. excluded from analysis because this small area is disjointed late data sets, in order to check whether the different groups from the rest of the country and has very distinct environ- presented similar patterns. Cells with fewer than five species mental conditions (Barbosa 1970). Therefore, the bushpig were excluded because preliminary analysis suggested that Potamochoerus porcus, occurring only in Cabinda, was this cut-off represented a good compromise between elimi- excluded from analysis. The African manatee Trichechus nating potential biases and the artefacts created by few senegalensis was also excluded because it is exclusively records per cell, and the need to retain a sufficiently large aquatic. The subspecies considered in the original publica- number of cells to obtain meaningful spatial patterns tions were maintained because they correspond to taxa that (Heikinheimo et al. 2007, Kreft & Jetz 2010). Dissimilarity could be easily recognized by field observers. In common between each pair of cells was estimated using the Simpson with other biogeographical and phylogeographical studies dissimilarity index (βsim) because it is independent of species (e.g. Lorenzen et al. 2012), we used the term ungulate in a richness gradients and does not take double absences into descriptive sense to refer to hoofed mammals, though it account (Koleff et al. 2003, Kreft & Jetz 2010). The cells were does not correspond to a well-defined taxonomic unit. then clustered by applying Ward’s minimum variance clus-

Nomenclature follows Wilson and Reeder (2005), unless tering method to the βsim dissimilarity matrix (Borcard et al. indicated otherwise (Table S1 in Appendix S1). 2011). To determine the optimal number of clusters, we examined fusion level plots of the merging height of nodes in the dendrogram against the number of clusters and Environmental data set applied the L-method (Salvador & Chan 2004). We evalu- Relationships between species assemblages and environ- ated the validity of cluster results by using the cophenetic mental conditions were explored by using climate, soil, correlation coefficient (Legendre & Legendre 1998, Kreft & landform and vegetation variables (Tables S2 and S3 in Jetz 2010). We also mapped the spatial distribution of the Appendix S1). Climate data were obtained from WorldClim groups to check for spatial coherence (i.e. the association of (version 1.4) at a resolution of 2.5 arc-min, using the 19 a group with a well-defined spatial region). When groups variables describing aspects of temperature and precipita- were not associated with well-defined regions, we reduced tion (Hijmans et al. 2005), and elevation. They encom- the number of groups to maximize spatial interpretability passed the period from 1950 to 2000 and were resampled to (Legendre & Legendre 1998). Analyses were implemented in mirror species data resolution. A principal component r v.3.0.1 (Anonymous 2005) using the ‘cluster’ (Maechler analysis was performed to investigate multicollinearity and et al. 2014) and ‘vegan’ (Oksanen et al. 2013) packages. reduce dimensionality (Dormann et al. 2013). The four To visualize the spatial pattern of biogeographical regions most uncorrelated climate variables were carried on to sub- obtained for carnivores, ungulates and rodents at each level sequent analysis (Table S2 in Appendix S1). The dominant of cluster analyses, we converted grid cells into a network of soil type and landform of each grid cell were extracted from Thiessen polygons (Schulman et al. 2007, Vale & Jenkins the Soil and Terrain Digital Database for Southern Africa vs. 2012). Each Thiessen polygon contained the centre of a 1.0 (SOTERSAF, available at http://www.isric.org/data/soil- single grid cell with species occurrence records (anchor and-terrain-database-southern-africa-ver-10-sotersaf). Soil point), and delimited the area within which any locality is categories were placed in major groups according to the closer to its anchor point than to the anchor point of all World Reference Base for Soil Resources (Anonymous other Thiessen polygons (Lo & Yeung 2002). Thiessen net- 2006). The dominant vegetation type was extracted from works provide a simple method to visualize spatial variation the phytogeographic map of Angola (Barbosa 1970). The in collection intensity because polygons tend to be larger total number of species records per grid cell was included to where sampling was sparser (Schulman et al. 2007, Vale & account for the possibility of sampling effort influencing Jenkins 2012). The average area of Thiessen polygons thus perceived variation in species assemblage patterns (Barbosa provided a surrogate of sampling effort in different biogeo- et al. 2010). graphical regions (Schulman et al. 2007, Vale & Jenkins 2012). Data analysis INDICATOR SPECIES IDENTIFICATION OF BIOGEOGRAPHICAL REGIONS Indicator species analysis was used to explore the composi- Data analysis was designed to identify biogeographical tion of mammal assemblages in the biogeographic regions regions, corresponding to sets of grid cells that are more and to identify which species discriminate the different similar in species composition to each other than to any regions well and can thus be interpreted as being character- other set of grid cells. Separate analyses were carried out for istic of them (Dufrêne & Legendre 1997, Proches¸& the overall data set, and for the rodent, carnivore and ungu- Ramdhani 2012). An indicator value (IndVal) was estimated

106 Mammal Review 45 (2015) 103–116 © 2015 The Mammal Society and John Wiley & Sons Ltd P. Rodrigues et al. Biogeographical regionalization of Angolan mammals for each species, as the product of the relative frequency of coherent geographical regions. Maps produced using the species in the target site group divided by the sum of smaller numbers of clusters suggested that maximum relative frequencies over all groups, and the relative fre- spatial interpretability was reached for subdivisions in three quency of occurrence of the species inside the target site or four clusters (Fig. 2). group. The maximum value (IndVal = 1) is given to a For the overall data set, the first split retrieved two well- species when it is found in all sites of a group (maximum defined regions, one in northern and central Angola, and specificity) and exclusively in that group (maximum fidel- another in the south and extreme east of the country ity). We considered a species to be an indicator when its (Fig. 2a). The second split (three clusters) largely retained IndVal > 0.50 for a P < 0.05. The statistical significance of the north-central region, while in the south, it separated the each IndVal was assessed though a randomization proce- Namibe region, roughly corresponding to the Kaokoveld dure with 999 permutations. Analysis was performed using desert and the Namibian dry savanna woodlands of the the function ‘multipatt’ from the r package ‘indicspecies’ south-western coastal plain, and the Cunene-Cuando (De Cáceres & Legendre 2009). Cubango region, mainly encompassing the Mopane and baikiaea woodlands along the border with Namibia. Spatial coherence was largely lost in the third and fourth splits, ENVIRONMENTAL RELATIONSHIPS though the Namibe and Cunene-Cuando Cubango regions Distance-based redundancy analysis (dbRDA) was used to retained their identity. The Central Plateau was divided into investigate the influences of environmental variables on the two or three loosely defined regions, which strongly species composition of mammalian assemblages, thus con- intergraded into each other. tributing to our interpretation of the results of the cluster Clusters identified using the ungulates shared some simi- analysis. The dbRDA technique uses redundancy analysis larities to those obtained with the overall data set, though but allows the analysis to be based on a range of ecologically with some differences (Fig. 2b). The first split also separated meaningful measures of dissimilarity, through the use of the northern from the southern regions, but in contrast to principal coordinate analysis (Legendre & Anderson 1999). the global analysis, part of the Central Plateau was assigned β Dissimilarity was based on sim to maintain consistency with to the southern region. The second split separated the the cluster analysis, but values were square-root trans- Namibe region from the rest of the southern region. The formed to eliminate negative eigenvalues from the principal third split further subdivided the southern region, separat- coordinate analysis (Legendre & Anderson 1999). We ing the Cunene-Cuando Cubango region from the southern started analysis by building a full model including the seven part of the Central Plateau. The northern-central region environmental variables and the surrogate of sampling was subdivided in the fourth split, largely separating the effort. A backward elimination procedure was then used to Congolian woodland-savanna mosaics of the north and remove the variables least related to variation in mamma- north-west, from the southern miombo woodlands of the lian assemblage. We eliminated all variables with P > 0.05, as Central Plateau. However, the two subregions did not show judged from Monte Carlo permutation tests (1000 permu- well-defined boundaries, intergrading to some extent into tations). The overall significance of the reduced model and each other. that of individual axes was also assessed with permutation Clustering for rodents and carnivores showed weak tests. All analyses were performed in r using ‘vegan’ spatial patterns (Fig. 2c, d). Nevertheless, the Namibe region (Oksanen et al. 2013). was clearly identified from the carnivore data set, which also showed some separation between the northern and southern regions of Angola (Fig. 2d). RESULTS

Biogeographical regions Indicator species Dendrograms produced in cluster analysis retained about For the overall data set and considering the division in three 40–50% of pairwise dissimilarity between grid cells, consid- broad regions (Fig. 2a), a total of 18 species (all ungulates) ering either the overall data set or each of the three species showed an IndVal > 0.50 for a P < 0.05 and were thus iden- groups (carnivores, ungulates and rodents; Table S4 in tified as indicators (Fig. 3a; Table S5 in Appendix S2). The Appendix S2). The numbers of clusters identified by the northern-central region was mostly associated with the L-method were five for carnivores, 11 for ungulates, and 12 presence of African forest buffalo Syncerus caffer nanus, for rodents and the overall data set (Fig. S3 in Appen- bushbuck Tragelaphus scriptus, bushpig Potamochoerus dix S2). However, visual inspection of maps suggested that larvatus and hippopotamus Hippopotamus amphibius.The the number of subdivisions identified was overly large as Namibe region was characterized primarily by the presence there was no clear matching between clusters of grids and of springbok Antidorcas marsupialis, Kirk’s dik-dik

Fig. 2. Sequence of clustering of the mammal data cells for the ﬁrst four splits of the cluster analysis and for each data set: (a) all species; (b) ungulates; (c) rodents; (d) carnivores. Colours are used to differentiate clusters within each image and do not imply one-to-one matches between images of the same data set.

Madoqua kirkii, klipspringer Oreotragus oreotragus, gems- Gerbilliscus validus (0.39) and Griselda’s Lemniscomys bok Oryx gazella, mountain zebra Equus zebra, greater kudu Lemniscomys griselda (0.38); Cunene-Cuando Cubango Tragelaphus strepsiceros,steenbokRaphicerus campestris and region: highveld gerbil Gerbilliscus brantsii (0.30), black- Burchell’s zebra Equus burchellii. The Cunene-Cuando tailed Thallomys Thallomys nigricauda and spotted hyena Cubango region was associated with blue wildebeest Crocuta crocuta; Namibe region: Cape fox Vulpes chama Connochaetes taurinus, roan antelope Hippotragus equinus, (0.44), Namaqua rock rat Aethomys namaquensis (0.44) and elephant Loxodonta spp., buffalo Syncerus caffer caffer, rock hyrax Procavia (capensis) welwitschii (0.42). common tsessebee Damaliscus lunatus and sable antelope Considering the four broad regions identiﬁed for ungu- Hippotragus niger niger. Albeit with much lower IndVals, lates (Fig. 2b), the species with strong IndVals for the there were some rodents and carnivores associated with the Namibe region were the same as those identiﬁed in the three regions, including for instance in the north-central overall analysis, though there were some differences in their region: Mastomys angolae (0.44), Southern savanna gerbil relative importance (Fig. 3b, Table S6 in Appendix S2).

(a) 0 10 20 30

Syncerus c. nanus (0.60) Tragelaphus scriptus (0.57) Region A Potamochoerus larvatus (0.56) Hippopotamus amphibius (0.51)

Antidorcas marsupialis (0.81) All species Madoqua kirkii (0.72) Oreotragus oreotragus (0.68) Oryx gazella (0.67) Equus zebra (0.66) Tragelaphus strepsiceros (0.57) Region C Raphicerus campestris (0.56) Equus burchellii ( 0.54)

ConnochaethesHippotragus equinus taurinus (0.62) (0.71) Loxodonta africana & L. cyclotis (0.62) Region B Syncerus caffer caffer (0.52) Damaliscus lunatus (0.52) Hippotragus niger niger (0.51)

Fig. 3. Dendrograms of the dissimilarity for the overall dataset (all species; a) and for the ungulate species dataset (b), and the biotic regions, built with Ward’s Minimum Variance clustering with βsim index of dissimilarity. Indi- cator species are shown for each biotic region, and their IndVal for each biotic region is given in parentheses. The top horizontal axis represents βsim dissimilarity values.

There were also similarities between analyses for the Environmental factors Cunene-Cuando Cubango, though there was increasing IndVal of the giraffe Giraffa camelopardalis and black rhino The dbRDA revealed a significant effect of environmental Diceros bicornis, while the roan antelope became one of the variables on assemblage patterns, which accounted for indicator species of the Central Plateau region. Other indi- about 24% of total variation in the dissimilarity matrix for cator species of the Central Plateau were oribi Ourebia the overall mammalian data set (Fig. 4a, Table S7 in Appen- ourebi, southern reedbuck Redunca arundinum, southern dix S2). The first axis was significant (F = 50.24, P < 0.005; lechwe Kobus leche and common eland Taurotragus oryx. 9% of total variation; 37% of constrained species– In the northern region, the main indicator species were environment variation), and it largely reflected a climatic forest buffalo, bushbuck, bushpig, yellow-backed duiker gradient of increasing mean temperature and precipitation, Cephalophus silvicultor and defassa waterbuck Kobus more densely forested vegetation types, and soils associated ellipsiprymnus defassa. For rodent and carnivore data sets, with more humid and forested conditions (ferralsols and no indicator analysis was performed given the weak results luvisols). The axis largely separated the drier and more open obtained in the cluster analysis. Namibe and Cunene-Cuando Cubango regions from the

Fig. 4. Triplots of distance-based redundancy analysis (dbRDA) for the overall dataset (all species; a) and for the ungulate species dataset (b). Ellipses represent the 95% confidence intervals of biogeographical regions identified in cluster analysis. Arrows indicate the gradients of quantitative environmental variables (TempMaxWM: maximum temperature of the warmest month; TempMeanDQ: mean temperature of driest quarter; PrecAnnual: annual precipitation; PrecWQ: precipitation of warmest quarter). Qualitative variables are represented by their centroids (codes in Appendix S1). Only the first three (for all species) and two (for ungulates) indicator species of each group are represented (species codes in Appendix S1).

110 Mammal Review 45 (2015) 103–116 © 2015 The Mammal Society and John Wiley & Sons Ltd P. Rodrigues et al. Biogeographical regionalization of Angolan mammals moister and more forested northern-central region. The Limitations and potential shortcomings second axis was also significant (F = 32.8, P < 0.005; 6% of total variation; 24% of constrained variation), and it mainly Although our study has some limitations and potential corresponded to a gradient of increasing maximum tem- shortcomings, these are unlikely to affect our key conclu- perature of the warmest month. Along this axis, there was a sions. In common with other studies based on biological separation of the two drier biogeographical regions. The surveys in remote regions (Ponder et al. 2001, Pyke & Namibe region was associated with desert and steppe veg- Ehrlich 2010), our study was affected by the concentration etation, hard rock, shallow and stony soils (leptosols), or of records in National Parks and Game Reserves, along arid zone soils (calcisols), and steep landforms. In contrast, roads and railway lines, near human settlements and farms, the Cunene-Cuando Cubango region was associated with and in otherwise more visited areas (Crawford-Cabral & higher temperature in the warmest months, and shrub and Mesquitela 1989). Also, observers ranged from professional tree savanna vegetation. zoologists to inexperienced travellers and hunters, which The species–environment relationships extracted by the may have biased the data in less explored regions towards dbRDA for ungulates were also significant, accounting for the most conspicuous species. This might have contributed about 32% of total variation in the dissimilarity matrix for towards enhancing the identification of biogeographical the ungulate data set (Fig. 4b, Table S8 in Appendix S2). The regions for which more data were available, particularly first axis was significant (F = 68.6, P < 0.005; 14% of total the Namibe and the Cunene-Cuando Cubango regions. variation; 43% of constrained species–environment varia- However, these regions are unlikely to be artefacts because tion), and it also highlighted a gradient stretching from the they are associated with very particular environmental con- drier Namibe and Cunene-Cuando Cubango regions to the ditions and with a set of mammalian species that are rare or moister and more forested northern region. The Central absent elsewhere in Angola (Crawford-Cabral & Veríssimo Plateau had an intermediate position along this axes, reflect- 2005). Moreover, the north–south division into two or three ing smooth north–south transitions rather than sharp dis- biogeographical regions is unlikely to be a product of biased continuities between regions. The second axis was also data, because it was associated with marked environmental significant (F = 40.89, P < 0.005; 8% of total variation; 26% gradients that are expected to influence the distributions of of constrained variation); it mainly separated the Namibe mammalian species. Including the number of records of region from the Cunene-Cuando Cubango region and was mammals per cell as a surrogate of sampling effort did not also defined by the gradient of increasing temperature in affect the results of dbRDA, further suggesting that sam- the warmest months identified for the overall assemblage. pling bias did not contribute significantly to the main pat- ternsweobserved. Another potential problem was that the overall data set DISCUSSION was dominated by ungulate data; far fewer records were available for rodents and carnivores. This probably explains We identified four main biogeographical units in Angola: the similarity of the results obtained in the overall analysis one northern region (Zaire-Lunda-Cuanza), one central and those obtained from the ungulate data sets, though the region (Central Plateau) and two regions in the south latter provided more detailed geographical patterns. It (Namibe and Cunene-Cuando Cubango). Although these cannot be ruled out, however, that the stronger associations regions emerged clearly when we considered ungulate dis- observed for ungulates resulted from their closer association tributions, the results for rodents and carnivores were to specific vegetation types than rodents and carnivores, largely inconclusive. The biogeographical patterns identified though further information would be needed to support were associated with a marked environmental gradient, this view. Overall, biogeographical regions should be identi- from more humid and forested areas in the north and fied by considering the ungulate data, pending further north-west, to drier and more open woodlands and savan- information to assess whether the results for ungulates nas in the south and south-east. Overall, our results are in apply generally to mammalian groups. line with the recent biogeographical regionalization of Africa proposed by Linder et al. (2012), though we suggest some refinements to the boundaries between regions and Biogeographical units add detail to regions previously recognized as largely homogeneous. Our results also agree with a recent worldwide Based on the results for ungulates (Fig. 4b), we suggest that analysis that places northern Angola in the transition Angola can be divided into four biogeographical regions, between Guineo-Congolian and South African regions, each characterized by a set of indicator species and associ- though it does not recognize the three regions identified by ated with particular environmental conditions and vegeta- our study within the latter region (Holt et al. 2013). tion types.

lechwe and the southern reedbuck also showed a high ZAIRE-LUNDA-CUANZA IndVal for this region, though they are widespread else- The Zaire-Lunda-Cuanza region corresponds to a rough tri- where in Angola. angle with vertices in the provinces of Zaire, Lunda Norte and Cuanza Sul, located in north and north-west Angola. It CUNENE-CUANDO CUBANGO emerged from the analysis of ungulates, while in the overall analysis, it was lumped with the Central Plateau. The region The Cunene-Cuando Cubango region was clearly identified matches the Angolan portion of the Shaba region identified in the analyses of both the overall and the ungulate data by Linder et al. (2012), which represents the transition sets. It encompasses a band running along the southern between Guineo-Congolian rain forests and the Zambezian border of Angola with Namibia, eastward of the Namibe region. It matches reasonably well to WWF Ecoregions, region and continuing farther north along the border with encompassing the blend of humid evergreen and semi- Zambia. The region is interrupted by a strip running along deciduous forests, woodlands, shrublands and grasslands the Cubango River and showing affinities with the Central characterizing the western Congolian forest-savanna mosaic Plateau. Although the overall biogeographical regionali- (Olson et al. 2001, Fig. 1). Along the coastal belt, it includes zation of Linder et al. (2012) includes the entire Cunene- most of the Angolan scarp savanna and woodlands. The Cuando Cubango region within the Zambezian region, in region also encompasses the northern part of the deciduous the partial analysis for mammals, it is included to a consid- broadleaf savannas and woodlands characterizing the erable extent within the South African/Kalahari region. The Angolan miombo woodlands. Two key indicator species Cunene-Cuando Cubango region therefore probably corre- were the forest buffalo and the yellow-backed duiker, which sponds to the northern limit of the Kalahari sub-region have their core range within the Congo Basin forests, reach- (Linder et al. 2012). In terms of WWF Ecoregions, most of ing their southern limit in Angola. Other indicators show the Cunene-Cuando Cubango region is included in the much the same pattern, including blue duiker Philantomba Zambezian baikiaea woodlands, though in the border with monticola, black-fronted duiker Cephalophus nigrifrons and the Namibe region, it encompasses the Angolan mopane bay duiker Cephalophus dorsalis. The other species with high woodlands. Most indicator species have a large natural dis- IndVal were bushpig, bushbuck and defassa waterbuck, tribution in savannahs of southern and eastern Africa, but which have a relatively wide distribution both in Sub- they have a restricted distribution elsewhere in Angola: Saharan Africa and Angola, but which, in this region, wildebeest, giraffe, sable antelope, black rhino, common occupy a band below the southern limit of the Congo Basin tsessebee, buffalo, hartebeest Alcelaphus buselaphus and forest, spreading from the coast into Zambia. These species impala Aepyceros melampus melampus. Elephants in Angola tend to be rare or absent in southern Angola. are mostly associated with the Cunene-Cuando Cubango region and are largely absent from the miombo woodlands, though they also occur in the Congolian forest-savanna CENTRAL PLATEAU mosaics (Crawford-Cabral & Veríssimo 2005). The Central Plateau region corresponds to the south and south-eastern parts of the central plateau of Angola. It inter- NAMIBE grades to some extent with the Zaire-Lunda-Cuanza region, suggesting that there is a north–south gradient in mamma- The Namibe region was identified by both the overall and lian assemblages, rather than two well-defined regions. This the ungulate data sets, and encompasses south-western view is supported by the lumping of the two regions in the Angola. It matches closely the south-western Angola region analysis using the overall data set. The Central Plateau is defined by Linder et al. (2012), which is part of the large included in the Zambezian region of Linder et al. (2012) Southern Africa Region. It has also a strong correspondence and encompasses to a large extent the WWF Ecoregion of with the WWF Ecoregions of the Kaokoveld desert, repre- the Angolan miombo woodlands (Olson et al. 2001). It senting the northern part of the vast Namib Desert, and the comprises moist, deciduous broadleaf savannas and wood- Namibian savanna woodlands. Rock hyrax Procavia lands dominated by the genera Brachystegia, Julbernardia (capensis) welwitschii and black-faced impala Aepyceros and Isoberlinia, interspersed with grassland areas. The main melampus petersi are characteristic of the region, being indicator species are widespread in Africa, occurring in the endemic to south-western Angola and north-western belt of woodlands and savannas surrounding the Guineo- Namibia. Yellow-spotted rock hyrax Heterohyrax brucei, Congolian region: oribi, roan antelope, elande, common Kirk’s dik-dik and, to a lesser extent, klipspringer also have a warthog Phacochoerus africanus and bush duiker Sylvicapra restricted distribution, though they also occur in arid and grimmia. Also characterizing this region is the local endemic semi-arid habitats in disjointed areas of north-eastern and giant sable antelope Hippotragus niger variani. The southern eastern Africa. Other indicators such as springbok, gemsbok

112 Mammal Review 45 (2015) 103–116 © 2015 The Mammal Society and John Wiley & Sons Ltd P. Rodrigues et al. Biogeographical regionalization of Angolan mammals and mountain zebra are associated with dry grasslands, are high and inaccessible (Frost 1996). Furthermore, many bushland and/or shrubland of south-western and southern savanna herbivores prefer areas of low tree density (Riginos Africa, showing a marginal extension into south-western & Grace 2008), which contributes to the absence of many Angola. Steenbok, greater kudu and Burchell’s zebra are species in dense miombo woodland. The region thus prob- widespread in eastern and southern Africa, but in Angola, ably represents a barrier that prevented the penetration they occur primarily in the south and south-west. southwards of Congolian species, and the penetration northwards of species from southern Africa. The coastal Namibe region seems to be somewhat eccen- Environmental influences on tric to the dominant north–south climate gradient (Le biotic regionalization Houérou 2009): it is an arid to hyper-arid tropical region The results of dbRDA revealed a north–south sequence of with relatively low temperature and little annual variation intergrading biogeographical regions which follows the in temperature, which is shaped by the local effects of the dominant climate gradients in this region of Africa (Le cold Benguela current. The region shares with western Houérou 2009), and the associated variation in soil and veg- Namibia a range of unique mammalian species adapted etation types. This sequence also matches closely a strong to desert conditions and favours also the occurrence of north–south gradient in closed canopy forest cover (Hansen other dry grassland and open savanna species. Recent et al. 2013) with a progressive southwards transition to phylogeographic studies suggest that this region was a Pleis- savannahs (Murphy & Bowman 2012). In the north and tocene refugium for a number of species, some of which north-west, the climate is predominantly humid equatorial, show genetic divergence from populations elsewhere in with high precipitation following a bimodal pattern, rain Africa (Lorenzen et al. 2012). concentrated in summer and temperature with low annual Although human activities are known to affect African amplitude (Le Houérou 2009). These conditions are associ- mammals strongly (Ceballos & Ehrlich 2002, Craigie et al. ated with dense evergreen forest and woodland mosaics 2010), it is unlikely that humans have shaped to a significant (Hansen et al. 2013), which allow the penetration of extent the large-scale biogeographical patterns observed in Congolian elements into the Zaire-Lunda-Cuanza region. this study. This is because the study was based on data col- The southern penetration of Congolian species seems to be lected from a long period before intense poaching resulted favoured also by the presence of dense gallery forests, which in major population declines and local extirpation of large may provide suitable habitat conditions for them in other- ungulates and carnivores in Angola (Crawford-Cabral & wise drier forest areas (Crawford-Cabral & Veríssimo 2005). Veríssimo 2005, Pitra et al. 2006, Chase & Griffin 2011). Southward, the climate progressively ranges from tropical Therefore, the data set probably reflects the natural distri- humid to semi-arid, with declining annual precipitation, bution of most species, as suggested for instance by the rela- and increases in the annual amplitude of temperature and tively large ranges observed for sensitive species such as in the temperature of the warmest month (Le Houérou elephants and black rhinos (Crawford-Cabral & Veríssimo 2009). Except for the desert region of Namibe, the driest 2005). Also, there is no evidence that the data set included extreme of the climate gradient is found in the Cunene- occurrences of species that had been translocated outside Cuando Cubango region, which is occupied by dry- their natural ranges, although translocation is known to deciduous savannas and allows the northward penetration have occurred in other regions of Africa (Spear & Chown of many species widespread elsewhere in southern Africa. 2009). Finally, it is unlikely that anthropogenic fires influ- The Central Plateau occupies an intermediate position in enced our results significantly, though they are widespread the north–south gradient, showing a set of a few character- in African ecosystems and are known to affect African istic species that are widespread in southern and eastern mammals strongly at the local and landscape scales (e.g. Africa, and also occur elsewhere in other regions of Angola. Bond & Archibald 2003). This view is supported by recent The only exception is the endemic giant sable antelope, research suggesting that the large-scale distribution of which has a very small natural range covering only a small forests and savannahs is mostly determined by climatic and proportion of the region. Therefore, the Central Plateau edaphic controls (Murphy & Bowman 2012, Zeng et al. appears to be separated from the other regions, mainly 2014), despite the influence of fire on local vegetation struc- because it is occupied at relatively low density by a few ture (e.g. Ryan & Williams 2011). Therefore, we believe that widespread species (Huntley 1974), rather than being char- the Angolan biogeographical regionalization described here acterized by a particular set of unique species. This may be a largely reflects the operation of natural rather than anthro- consequence, at least partly, of the poor browsing condi- pogenic processes and thus, it provides a baseline for assess- tions provided by the dry miombo woodlands covering this ing changes induced by relatively recent impacts such as region: there is a shortage of browse in the dry season, the poaching (Crawford-Cabral & Veríssimo 2005) and defores- quality of leaf matter is low, and the canopies of most trees tation (Hansen et al. 2013).

Overall, our results suggest that the biogeographical pat- Borcard D, Gillet F, Legendre P (2011) Numerical Ecology with terns observed in this study are shaped to a large extent by R. Springer, New York, USA. the position of the country at the crossroads between the Carcasson RH (1964) A preliminary survey of the zoogeography humid, evergreen forests of the Congolian Basin, the desert of African butterflies. East African Wildlife Journal 2: 122–157. and arid savannas of south-western Africa, and the dry Ceballos G, Ehrlich PR (2002) Mammal population losses and deciduous woodlands and savannahs of southern and the extinction crisis. Science 296: 904–907. eastern Africa (Le Houérou 2009). Each of these habitat Chapin JP (1932) The birds of the Belgian Congo. I Bulletin of types is associated with a characteristic set of mammalian the American Museum of Natural History 65: 1–756. species that penetrate into Angola and intergrade into each Chase MJ, Griffin CR (2011) Elephants of south-east Angola in war and peace: their decline, re-colonization and recent other to a variable extent, probably depending on the domi- status. African Journal of Ecology 49: 353–361. nant vegetation types, which in turn are shaped by climate Craigie ID, Baillie JE, Balmford A, Carbone C, Collen B, Green and edaphic factors. It is likely that the same gradients also RE, Hutton JM (2010) Large mammal population declines in affect mammalian groups not analysed here, as well as other Africa’s protected areas. Biological Conservation 143: plant and animal groups (Linder et al. 2012). Analysing the 2221–2228. consistency of the patterns observed for a wide range of Crawford-Cabral J (ed) (1998) The Angola rodents of the species would be a valuable subject for further research. superfamily Muroidea. An account on their distribution. In: Such research could contribute greatly towards conservation Estudos Ensaios e Documentos, 161, 222. Instituto de planning in Angola and towards our understanding of Investigação Científica Tropical, Lisboa, Portugal. African biogeography. Crawford-Cabral J, Mesquitela LM (eds) (1989) Índice toponímico de colheitas zoológicas em Angola. In: Estudos Ensaios e Documentos, 151, 206. Instituto de Investigação ACKNOWLEDGEMENTS Científica Tropical, Lisboa, Portugal. This work was carried out in the scope of a partnership, Crawford-Cabral J, Simões AP (1987) Distributional data and funded by the Portuguese Science Foundation through the notes on Angolan carnivores (Mammalia: Carnivora) I – fellowship SFRH/BI/51642/2011, EDP Biodiversity Chair, Small and medium-sized species. Garcia de Orta, Série de Spanish National Research Council and EC IC&DT Call No Zoologia 14: 3–27. 1/SAESCTN/ALENT-07-0224-FEDER-001755. Thanks are Crawford-Cabral J, Simões AP (1988) Distributional data and notes on Angolan carnivores (Mammalia: Carnivora) due to J. Crawford-Cabral, Luís N. Veríssimo, François II – Larger species. Garcia de Orta, Série de Zoologia 15: Guilhaumon, A. Márcia Barbosa, Miguel Porto and Hugo 9–20. Rebelo for providing help in assembling information, Crawford-Cabral J, Veríssimo LN (eds) (2005) The ungulate analysing data and interpreting results. This is scientific fauna of Angola. Systematic list, distribution maps, database paper no. 2 from the Portuguese-Angolan TwinLab estab- report. In: Estudos Ensaios e Documentos, 163, 277. Instituto lished between CIBIO/InBIO and ISCED/Huíla, Lubango. de Investigação Científica Tropical, Lisboa, Portugal. de Klerk HMD, Crowe TM, Fjeldså J, Burgess ND (2002) REFERENCES Biogeographical patterns of endemic terrestrial Afrotropical birds. Diversity and Distributions 8: 147–162. Anonymous (2005) R: A Language and Environment for De Cáceres M, Legendre P (2009) Associations between species Statistical Computing. Version 3.0.1. R foundation for and groups of sites: indices and statistical inference. http://cran.r Statistical Computing, Vienna, Austria. -project.org/web/packages/indicspecies/index.html. http://cran.r-project.org/ Dormann CF, Elith J, Bacher S, Buchmann C, Carl G, Carré G Anonymous (2006) World Reference Base for Soil Resources 2006. et al. (2013) Collinearity: a review of methods to deal with it World Soil Resources Reports No. 103. FAO, Rome, Italy. and a simulation study evaluating their performance. Anonymous (2013) IUCN Red List of Threatened Species, version Ecography 36: 27–46. 2013.1. http://www.iucnredlist.org. Dufrêne M, Legendre P (1997) Species assemblages and Barbosa AM, Fontaneto D, Marini L, Pautasso M (2010) Is the indicator species: the need for a flexible asymmetrical human population a large-scale indicator of the species approach. Ecological Monographs 67: 345–366. richness of ground beetles? Animal Conservation 13: Figueiredo E, Smith GF, César J (2009) The flora of Angola: first 432–441. record of diversity and endemism. Taxon 58: 233–236. Barbosa LAG (1970) Phytogeographic Map of Angola. Instituto de Frost P (1996) The ecology of miombo woodlands. In: Campbell Investigação Científica de Angola, Luanda, Angola. B(ed)The Miombo in Transition: Woodlands and Welfare in Bond WJ, Archibald S (2003) Confronting complexity: fire Africa, 11–57. CFIOR, Bogor, Indonesia. policy choices in South African savanna parks. International Gelderblom CM, Bronner GN, Lombard AT, Taylor PJ (1995) Journal of Wildland Fire 12: 381–389. Patterns of distribution and current protection status of the

Carnivora, Chiroptera and Insectivora in South Africa. South 1.15.2. http://cran.r-project.org/web/packages/ African Journal of Zoology 30: 103–114. cluster/index.html. Grubb P, Sandrock O, Kullmer O, Kaiser TM, Schrenk F (2000) Mills MSL (2010) Angola’s central scarp forests: patterns of bird Relationships between Eastern and Southern African mammal diversity and conservation threats. Biodiversity and faunas. In: Bromage T, Schrenk F (eds) African Biogeography, Conservation 19: 1883–1903. Climate Change and Early Hominid Evolution, 253–267. Monod T (1957) Les Grands Divisions Chronologiques de Oxford University Press, New York, USA. l’Afrique. C.S.A./C.C.T.A. Publ. No. 24,1–150. C.S.A./C.C.T.A., Hansen MC, Potapov PV, Moore R, Hancher M, Turubanova London, UK. SA, Tyukavina A et al. (2013) High-resolution global maps of Muñoz AR, Real R, Olivero R, Márquez AL, Guerrero JC, 21st-century forest cover change. Science 342(6160): 850–853. Bárcena SB, Vargas JM (2003) Biogeographical zonation of Heikinheimo H, Fortelius M, Eronen J, Mannila J (2007) African hornbills and their biotic and geographic Biogeography of European land mammals shows characterizations. Ostrich 74: 39–47. environmentally distinct and spatially coherent clusters. Murphy BP, Bowman DMJS (2012) What controls the Journal of Biogeography 34: 1053–1064. distribution of tropical forest and savanna? Ecology Letters 15: Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005) 748–758. Very high resolution interpolated climate surfaces for global Myers N, Mittermeier RA, Mittermeier CG, Da Fonseca GA, land areas. International Journal of Climatology 25: 1965–1978. Kent J (2000) Biodiversity hotspots for conservation Holt BG, Lessard JP, Borregaard MK, Fritz SA, Araújo MB, priorities. Nature 403: 853–858. Dimitrov D et al. (2013) An update of Wallace’s Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, zoogeographic regions of the world. Science 339(6115): O’Hara RB, Simpson GL, Solymos P, Stevens MHH, Wagner 74–78. H (2013) Vegan: community ecology package, version 2.0–8. Huntley BJ (1974) Outlines of wildlife conservation in Angola. http://cran.r-project.org/web/packages/vegan/index.html. Journal of the Southern African Wildlife Management Olson DM, Dinerstein E, Wikramanayake ED, Burgess ND, Association 4: 157–166. Powell GVN, Underwood EC et al. (2001) Terrestrial Koleff P, Gaston KJ, Lennon J (2003) Measuring beta diversity ecoregions of the world: a new map of life on Earth. for presence-absence data. Journal of Animal Ecology 72: Bioscience 51: 933–938. 367–382. Penner J, Wegmann M, Hillers A, Schmidt M, Rödel MO (2011) Kreft H, Jetz W (2010) A framework for delineating A hotspot revisited–abiogeographical analysis of West biogeographical regions based on species distributions. African amphibians. Diversity and Distributions 17: Journal of Biogeography 37: 2029–2053. 1077–1088. Ladle RJ, Whittaker RJ (2011) Conservation Biogeography. Pitra C, Vaz Pinto P, O’keefe B, Munro S, Van Vuuren B, Blackwell Publishing, Oxford, UK. Robinson T (2006) DNA-led rediscovery of the giant sable Le Houérou HN (2009) Bioclimatology and Biogeography of antelope in Angola. European Journal of Wildlife Research 52: Africa. Springer, Heidelberg, Germany. 145–152. Lebrun J (1947) La végétation de la plaine alluviale au sud du Ponder WF, Carter GA, Flemons P, Chapman RR (2001) lac Édouard. Exploration du Parc National Albert. Intitut des Evaluation of museum collections data for use in Parcs Nationaux du Congo Belge. Bruxelles. Fascicule 1, biodiversity assessment. Conservation Biology 15: 1–800. 648–657. Legendre P, Anderson MJ (1999) Distance-based redundancy Poynton JC (1964) The Amphibia of southern Africa. Annals of analysis: testing multispecies responses in multifactorial the Natal Museum 17: 1–334. ecological experiments. Ecological Monographs 69: 1–24. Proches¸ S, Ramdhani S (2012) The world’s zoogeographical Legendre P, Legendre L (1998) Numerical Ecology, 2nd English regions conﬁrmed by cross-taxon analyses. Bioscience 62: ed. Elsevier Science B.V., Amsterdam, The Netherlands. 260–270. Linder HP, de Klerk HM, Born J, Burgess ND, Fjeldså J, Rahbek Pyke GH, Ehrlich PR (2010) Biological collections and C (2012) The partitioning of Africa: statistically deﬁned ecological/environmental research: a review, some biogeographical regions in sub-Saharan Africa. Journal of observations and a look to the future. Biological Reviews 85: Biogeography 39: 1189–1205. 247–266. Lo CP, Yeung AKW (2002) Concepts and Techniques of Riginos C, Grace JB (2008) Savanna tree density, herbivores, and Geographic Information Systems. Prentice Hall, Upper Saddle the herbaceous community: bottom-up vs. top-down effects. River, New Jersey, USA. Ecology 89: 2228–2238. Lorenzen ED, Heller R, Siegismund HR (2012) Comparative Romeiras MM, Figueira R, Duarte MC, Beja P, Darbyshire I phylogeography of African savannah ungulates. Molecular (2014) Documenting biogeographical patterns of African Ecology 21: 3656–3670. timber species using herbarium records: a conservation Maechler M, Rousseeuw P, Struyf A, Hubert M, Hornik K perspective based on native trees of Angola. PLoS ONE 9(7): (2014) Cluster: cluster analysis basics and extensions, version e103403.

Rowe-Rowe DT, Taylor PJ (1996) Distribution patterns of White F (1965) The savanna-woodlands of the Zambezian and terrestrial mammals in KwaZulu-Natal. South Africa Journal of Sudanian domains. Webbia 19: 651–681. Zoology 31: 131–144. Wilson DE, Reeder DM (2005) Mammal Species of the World: a Ryan CM, Williams M (2011) How does ﬁre intensity and Taxonomic and Geographic Reference, 3rd ed. The John frequency affect miombo woodland tree populations and Hopkins University Press, Baltimore, USA. biomass? Ecological Applications 21: 48–60. Zeng Z, Chen A, Piao S, Rabin S, Shen Z (2014) Environmental Salvador S, Chan P (2004) Determining the number of determinants of tropical forest and savanna distribution: a clusters/segments in hierarchical clustering/segmentation quantitative model evaluation and its implication. Journal of algorithms. Proceedings of the 16th IEEE – International Geophysical Research: Biogeosciences 119: 1432–1445. Conference on Tools with Artiﬁcial Intelligence, 576–584. Institute of Electrical and Electronics Engineers, Piscataway, New Jersey, USA. SUPPORTING INFORMATION Schulman L, Toivonen T, Ruokolainen K (2007) Analysing botanical collecting effort in Amazonia and correcting for it Additional supporting information may be found in the in species range estimation. Journal of Biogeography 34: online version of this article at the publisher’s web-site. 1388–1399. Spear D, Chown SL (2009) The extent and impacts of ungulate Appendix S1. Supplementary materials and methods – lists translocations: South Africa in a global context. Biological of species and environmental correlates. Conservation 142: 353–363. Appendix S2. Supplementary results. Vale MM, Jenkins CN (2012) Across-taxa incongruence in patterns of collecting bias. Journal of Biogeography 39: 1744–1748.